AI
Put your cache breakpoint on the last block that never changes
Why prompt caching silently fails when your breakpoint sits on a timestamp or per-request block, and where to move it instead.
Read →Tips & Tricks · 4
Tips & Tricks
The small optimizations, settings, and ways of working I lean on day to day — the minute things that compound.
How a tip reaches the site
Tooling
Gate your admin routes — then re-verify the token in your own code
Putting a login in front of an endpoint is one lock. Having the endpoint check the proof itself is the second. Defense in depth is cheap when you wire it once.
Read → AI
On Sonnet 5, turn thinking OFF for structured extraction
Newer models default to adaptive thinking, which quietly eats your token budget. For 'pull these fields out of this text' work, disable it and set effort low.
Read → Optimization
Put a hard monthly cap on every AI API key
The five-minute setting that turns a runaway loop or a leaked key from a five-figure surprise into a capped, survivable one.
Read →